Adding application awareness in flexible optical networking Presenter: Dimitrios Klonidis Contributors: I. Tomkos, P. Khodashenas, C. Kachris Networks and Optical Communications group NOC
Outline Evolution of optical communication systems & networks What is flexible optical networking Application awareness in flexible optical networking the ACINO project approach Summary & Conclusions 2
Evolution of optical communication systems & networks 3
New services and network requirements Upcoming services are huge bandwidth generating sources: Video services Data services to/from mobile users Cloud services These new traffic sources lead to new characteristics: Rapidly changing traffic patterns - high traffic churn High peak-to-average traffic ratio Large data-chunk transfers Asymmetric traffic between nodes Increasing high-qos traffic Over-provisioning Over-provisioning Maintaining network resource overprovisioning is not possible any more 4 Source: Transmode
Bit Rate Distance Product (Gbit/s.Mm) Total Fibre Capacity (Tbit/s) Optical communications system capacity and bit-rate x distance product WDM Data from Prof. Andrew Ellis (Aston Uni) 10,000,000 TDM 1000 1,000,000 OFDM/CoWDM 100 Coherent Detection 100,000 Spatial Multiplexing 10 10,000 Total capacity 1 1,000 0.1 100 0.01 10 0.001 1 0.0001 0 1983 1987 1991 1995 1999 2003 2007 2011 2015 Year Published 0.00001 Traffic increases at a rate of 20-40% per year, while capacity of deployed SMF-based networks approaches fundamental limits Fiber bandwidth was consider for many years as an abundant resource, but we have almost utilized to the maximum extend the EDFA amplifiers bandwidth (i.e. while approaching the fundamental SE limits) A forward-looking option: deploy new fibers (or use strands of available SMF fibers) that can support multi-cores or/and multi-modes per core (SDM/Spatially-flexible networks) 5 A short-term solution: utilize the available fiber spectrum more efficiently - (Spectrally flexible networks)
The technology shift to coherent mesh networks Multi-level (>2) modulation formats for improved spectral efficiency and reach Coherent detection Polarisation division multiplexing/demultiplexing Digital signal pre/post processing algorithms High speed ASICs/FPGAs 10G 10G 100G 40G 400G Multi-rate variable spectrum interface 43Gb/s @50GHz 112Gb/s @50GHz 224Gb/s @50GHz 448Gb/s @80GHz 1Tb/s @170GHz Photonic ICs 6
What is flexible optical networking 7
Flexible/elastic optical networks The definition Flexible (Elastic) Optical Network = A network able to adapt its resources (link capacity, transmission bandwidth per node, switching capacity etc.) according to the connectivity (i.e. traffic) demands in an automated fashion Flexible, elastic, tunable, gridless, or adaptive are few examples of the terms used by the research community to describe such solutions The target Offer optimization of the poorly-filled wavelengths of the fixed spectrum grid via a flexible spectrum allocation that requires a new wavelength-grid enabling adaptive sub-wavelength and super-wavelength services Spectrum- Flexible Optical Node Flexible Spectrum f Spectrum- Flexible Optical Node Bandwidth-Flexible WSS 10G 40G 400G Bandwidth-Flexible WSS Adaptive Line-Rate/ Modulation Format Bandwidth-Flexible WSS 100G 400G 40G Bandwidth-Flexible WSS 8
Enabling elements to realize flexible networking Main building blocks for enabling flexible networks: Flexible transponders Flexible switching nodes Flexible Optical Networking Network planning & control plane 9
The Concept of Flexible Optical Networking 10
Flexible data transport options Optical OFDM Nyquist WDM Min. WSS resolution e.g. 12.5GHz or 11 m-qam (format selectable) Elctronic OFDM (i.e. with SC generated in el. Domain) Nyquist-FDM (i.e. with N- shaped spectra generated electronically)
Multiplexing schemes and modulation formats to compose super-channels Source: EU project FOX-C 12
A flexible node architecture Source: EU project FOX-C 13
Expanding the link capacity limits Time: Too expensive to go to higher symbol rates (>32Gboud) PMD starts becoming an issue Format level: Multi-level schemes have reduced transmission distance and increased processing complexity Polarization level: Is limited to two levels Frequency: Increased spectrally-efficiency with the use of flexible multiplexing schemes (i.e. Nyquist WDM and optical OFDM), BUT still limited capacity increase (Shannon limit) Expansion to other wavelength bands beyond C+L may be a solution (still limited), but requires wideband modules Space: This is the obvious yet unexplored dimension if capacity is limited then offer multiple systems in parallel BUT by simply increasing the number of systems, the cost and power consumption also increase linearly! 14
Hero transmission experiments 15 BUT all these are very good for the spatial capacity increase in Point-to- Point systems WHAT ABOUT using the spatial dimension for optical networking
Capacity expansion In Space The true use of the space domain requires the spatial integration of system elements * Significant efforts in the development of FMF and MCF (fibre integration) Multi-link amplification systems have also be proposed and developed Tx/Rx integration is a hot and very active topic BUT, all these are good only for the spatial capacity increase in Point-to-Point systems WHAT ABOUT using the spatial dimension for optical networking? * Peter J. Winzer, Spatial Multiplexing: The next frontier in network capacity scaling, Tutorial paper at ECOC 2013 16
The INSPACE channel allocation concept SMF- Bundle or FMF or MCF Fibre, Mode,Core N is the channel number (=7 in this example) N-WDM (a) (b) (N-1)/T Conventional optical OFDM Frequency (N-1)/2T Optical fast OFDM Frequency or OFDM or SC-M-QAM Degrees of Flexibility Modes/Cores Wavelengths Data rate (Modulation level) f f f f f Modes or Cores Spatial expansion of the spectrum over multiple modes/cores and therefore definition of a superchannel over two dimensions (instead of the spectrum only dimension) 18 : end-to-end allocated channel See: D.Klonidis, et al., IEEE Comm.Mag., Vol.53, no. 2, pp.69-78, Feb2015
Routing and resource optimization Traditional approaches use the wavelength channel allocation option Requires wavelength routing algorithms In flexible networks a number of subcarrier slots are now to be assigned Routing and spectrum allocation algorithms (RSA). As the modulation level can be selected on a connection basis, constraints tying it to the required bit-rate versus the achieved reach, are necessary. Routing modulation level and spectrum allocation algorithms (RMLSA). A further addition is the allocation of spatial super-channels to fiber cores/modes Spatial routing modulation level and spectrum allocation algorithms (S-RMLSA) The discussed algorithms can address the offline network planning phase, or they can be applied to dynamically provision connection requests. Speed??? 19 See: I. Tomkos, et al. Proceedings of the IEEE, vol.102, no.9, pp.1317-1337, Sept.2014.
Typical resource optimization process 20
Networking and Control 21 Significant improvements are needed in the control plane architecture that should provide a new set of functionalities, such as: Flexible/elastic traffic support Physical layer awareness Multi-domain and multi-vendor support Network virtualization Evolution towards cognitive networks
Networking and Control Spatial allocation of VN segments What does the space dimension bring in networking: More routing and channel allocation options more optimization options in terms of: Link capacity Cost (Capex-Opex) Energy efficiency Hitless spectral defragmentation using the space domain Support of actual network virtualization with spatial separation virtual network segments Capacity flexibility can be maintained in spectrum domain 22 BUT increased complexity in the control plane
Control plane solutions Space dimension Channel allocation options Control plane options for flexible networking legacy networks future flexible networks Spectrum dimension Spectrum + BW flex dimension Spectrum + BW + Spatially flex Fixed rate channel allocation in wavelength domain Adaptive rate channel allocation in wavelength domain Adaptive rate channel allocation in wavelength AND space domains λ reconfigurability, ROADMs Programmable (i.e. adaptive) devices Sliceable spectrum Extend flexibility (programmability) to space domain (switching nodes) Static spatial expansion to multiple links Static spatial expansion to multiple links Adaptive spatial expansion to multiple links/cores GMPLS based control plane Well defined and worked fine GMPLS extensions required to take the BW flexible option into consideration SDN concept into networking efficient control of programmable resources in a sliced spectrum A GMPLS may lead to suboptimal allocation of multi-dimension resources (+scalability issues) SDN can make the difference! Support of multi-dimensional flexibility and virtualization 23
Why SDN in flexible optical networks? SDN and GMPLS (the common centralized version) have the same goal the efficient routing of demands in a network. They both consider all the network resources available and can potentially optimize their use They both rely on the use of path computation to identify the less costly path However what changes are: a) The capabilities of the modern data plane (i.e. the DSP enabled hardware capabilities) b) The amount of data required to be optimized (processed) The key benefits of SDN in flexible optical networking are: The more efficient control and management of complicated networks The higher the complexity the more efficient (compared to the limitations of GMPLS) The manipulation of traffic (packet) according to type Extra capabilities when the IP type of traffic comes into play The capability to deal with virtualized network infrastructures for any potential future use (in case that an infrastructure sharing model is promoted) 24
Application awareness in flexible optical networking the ACINO project approach 26
The view of ACINO project So far, the optimization of the Flexible optical network resources is equivalent to the optimization of the data plane resources according to the end-to-end demands However, demands are generated from a diverse set of applications/services with different needs While the applications can be treated differently by the IP layer (QoS), they are aggregated and treated commonly before they are routed in the optical layer Concept 1: Since flexible optical networking can adapt to the end-to-end demands, why not to be adaptive to the type of services/applications per demand. SDN allows application control over the use of network resources, but this is again limited to the IP layer Concept 2: The need for an application-centric network orchestration is foreseen. 27
Approach: From application-unaware baseline Application-unaware baseline application classes could be treated differently at the IP layer by its built-in QoS mechanisms not currently a feature of the optical layer All traffic served by an IP interface mapped into an optical connection sent towards the destination IP interface DC A IP/Optical Transport Network D N2 Legend B N1 SD SD Application class red traffic from S to D Application class green traffic from S to D N3 SD Application class blue traffic from S to D DC C At the edge of the transport network traffic for all applications to the same next IP hop are translated into an optical wavelength E 29
Approach: to application-centricity! Application-centric networking keep the different application classes separate down to the optical layer Different service (latency, survivability, security,...) for apps IP/Opt: global vision, joint/differentiated optimization (virtual IP/Optical networks per class) DC A IP/Optical Transport Network D N2 Legend B N1 SD SD Application class red traffic from S to D Application class green traffic from S to D N3 SD Application class blue traffic from S to D 30 DC C At the edge of the transport network same/equivalent applications data are multiplexed into smaller optical sub-wavelengths (thanks to sliceable transponders) E
Orchestrator modules Dynamic Resource Allocation Online Planning primitives primitives primitives Approach: dynamic controller and primitives A dynamic IP/Optical orchestrator able to expose to applications primitives to be mapped into service Applications / Network Management System APIs Application-centric algorithms. Specific application class needs Chosen Primitives In-Operation Planning Yes Each 30 Minutes Objective: save energy. No PRIMITIVES SPACE. Routing metrics Bandwidth Latency. Survivability Optical Protection Multi-layer Restoration Controller logic. SERVICE (how network is programmed) N2 N1 N3 31
Typical IP and Optical layer optimization approaches A. R1 R1-R2 flow (60) R2 R3 R3-R1 flow (80) R3-R2 flow (20) Separate IP-Optical layer optimization Optimization of the IP Layer Shortest path routing Flexible OADM Flexible OADM Flexible OADM Flexible OADM Flexible OADM Flexible OADM Dimensioning of the optical Layer Sum { Max( IP Demand i ) } Mapping of IP requests over the optical layer topology B. R1-R2 flow (60) R1 R3 R3-R1 flow (80) R3-R2 flow (20) Joint IP-Optical layer optimization Joint dimensioning of IP links and optical circuits Flexible OADM R2 Flexible OADM Shortest IP/Optical path routing Flexible OADM Apply constraints for optimizing spectral (and/or spatial) resource utilization 32
IP Optical layer optimization with application awareness 33
Summary & Conclusions 34
Summary and conclusions Elastic/Flexible optical networking emerges from the need to optimize the network resources according to the endto-end demands, thus avoiding overprovision is assisted by major technology advancements in spectrally and data rate adaptive transceivers and switching elements generates new opportunities with respect to network planning and resource optimization options however the current research focus is restricted on the joint optimization of the IP and optical layer ignoring the characteristics of the application layer. Application awareness in elastic optical networks is introduced: by separating the IP demands into Application Classes with different end-toend routing characteristics by applying simultaneously different optimization criteria in the optical layer (driven by the AppClass demands) by providing new SDN orchestrator solutions that take into consideration all the different optimization dimensions and interface with both the IP ports and the optical switching nodes. 35
Thank you! For more information please visit: http://www.acino.eu/ for http://www.ict-fox-c.eu/ for http://www.ict-inspace.eu/ for or contact Dimitrios Klonidis: dikl@ait.gr Acknowledgement Dr. Domenico Siracusa from CREATE-NET for ACINO project the NOC group members: Ioannis Tomkos, Pouria Khodashenas, Christopher Kachris and Jose M. Rivas 36